Patient-Specific Computational Models of Coronary Arteries Using Monoplane X-Ray Angiograms

Using temporal and structural data to reconstruct 3D cerebral vasculature from a pair of 2D digital subtraction angiography sequences

PURPOSE

The purpose of this work is to present a new method for reconstructing patient-specific three-dimensional (3D) vasculature of the brain from a pair of digital subtraction angiography (DSA) image sequences from different viewpoints, e.g., from bi-plane angiography. Our long-term goal is to provide high resolution visualization of 3D vasculature with dynamic flow of contrast agent from limited data that is readily available during surgical procedures. The proposed method is the second of a three-stage process composed of 1) augmenting vessel segmentation with vessel radii and timing of the arrival of a bolus of contrast agent, 2) reconstructing a volumetric representation of the augmented vessel data from the augmented 2D segmentations, and 3) generating a 3D model of vessels and flow of contrast agent from the volumetric reconstruction. Unlike previous methods, which are either limited to relatively simple vessel structures or rely on multiple views and/or prior models of the vasculature, our method requires only a single pair of 2D DSA sequences taken from different view directions.

METHODS

We developed a new mathematical algorithm that augments vessel centerlines with vessel radii and bolus arrival times derived directly from the 2D DSA sequences to constrain the 3D reconstruction. We validated this method on digital phantoms derived from clinical data and from fractal models of branching tree structures.

RESULTS

In standard reconstruction methods, reconstruction by projection of two views into 3D space results in 'ghosting' artifacts, i.e., false 3D structure that occurs where vessels or vessel segments overlap in the 2D images. For the complex vascular of the brain, this ghosting is severe and is a major hurdle for methods that attempt to generate 3D structure from 2D images. We show that our approach reduces ghosting by up to 99% in digital phantoms derived from clinical data.

CONCLUSION

Our dramatic reduction in ghosting artifacts in 3D reconstructions from a pair of 2D image sequences is an important step towards generating high resolution 3D vasculature with dynamic flow information from a single DSA sequence acquired using bi-plane angiography.

Vessel segmentation for X-ray coronary angiography using ensemble methods with deep learning and filter-based features

BACKGROUND

Automated segmentation of coronary arteries is a crucial step for computer-aided coronary artery disease (CAD) diagnosis and treatment planning. Correct delineation of the coronary artery is challenging in X-ray coronary angiography (XCA) due to the low signal-to-noise ratio and confounding background structures.

METHODS

A novel ensemble framework for coronary artery segmentation in XCA images is proposed, which utilizes deep learning and filter-based features to construct models using the gradient boosting decision tree (GBDT) and deep forest classifiers. The proposed method was trained and tested on 130 XCA images. For each pixel of interest in the XCA images, a 37-dimensional feature vector was constructed based on (1) the statistics of multi-scale filtering responses in the morphological, spatial, and frequency domains; and (2) the feature maps obtained from trained deep neural networks. The performance of these models was compared with those of common deep neural networks on metrics including precision, sensitivity, specificity, F1 score, AUROC (the area under the receiver operating characteristic curve), and IoU (intersection over union).

RESULTS

With hybrid under-sampling methods, the best performing GBDT model achieved a mean F1 score of 0.874, AUROC of 0.947, sensitivity of 0.902, and specificity of 0.992; while the best performing deep forest model obtained a mean F1 score of 0.867, AUROC of 0.95, sensitivity of 0.867, and specificity of 0.993. Compared with the evaluated deep neural networks, both models had better or comparable performance for all evaluated metrics with lower standard deviations over the test images.

CONCLUSIONS

The proposed feature-based ensemble method outperformed common deep convolutional neural networks in most performance metrics while yielding more consistent results. Such a method can be used to facilitate the assessment of stenosis and improve the quality of care in patients with CAD.